CLAUDE.md, add Claude to code review on GitHub, and measure time to merge, not lines of code. Budget $150-250 per developer a month in usage on top of seats (Anthropic).| Week | Goal | Done when |
|---|---|---|
| 1 | Buy seats, pick 5-10 pilot engineers, record a baseline, lock down settings | Everyone in the pilot can run Claude Code, and nobody can read .env |
| 2 | Teach Claude your codebase | Each pilot repo has a CLAUDE.md under 200 lines and 2-3 shared skills |
| 3 | Add AI review and scheduled jobs, turn on metrics | @claude answers in PRs, 1-2 jobs run on a schedule, usage shows in your dashboard |
| 4 | Decide and expand | You've compared week 4 to the baseline and made a go or no-go call |
| Plan | Price per seat | Best for |
|---|---|---|
| Team, standard seat | $20/month billed annually, $25 monthly | PMs, designers and others who mostly use Claude chat |
| Team, premium seat | $100/month billed annually, $125 monthly, with 5x the usage of a standard seat | Engineers using Claude Code every day, on teams of 2-150 |
| Enterprise | $20/month billed annually, plus usage at API rates, 20-seat minimum | Teams that need SCIM, audit logs, IP allowlisting or custom data retention |
| OS | Path |
|---|---|
| macOS | /Library/Application Support/ClaudeCode/managed-settings.json |
| Linux and WSL | /etc/claude-code/managed-settings.json |
| Windows | C:\Program Files\ClaudeCode\managed-settings.json |
Loading editor...
{
"permissions": {
"allow": [
"Bash(npm run lint)",
"Bash(npm run test *)"
],
"deny": [
"Read(./.env)",
"Read(./.env.*)",
"Read(./secrets/**)"
],
"disableBypassPermissionsMode": "disable"
}
}"allowManagedPermissionRulesOnly": true..claude/settings.local.json, .claude/settings.json in the repo, then ~/.claude/settings.json per user. Lists merge across files instead of replacing each other (docs).CLAUDE.md is the file Claude reads at the start of every session. It's the difference between an agent that guesses your conventions and one that follows them.| Level | Where | Who it's for |
|---|---|---|
| Organization | /Library/Application Support/ClaudeCode/CLAUDE.md (macOS), /etc/claude-code/CLAUDE.md (Linux) | Rules for every repo: security, review, commit style |
| Project | ./CLAUDE.md in the repo, committed | Everyone working on this repo |
| Local | ./CLAUDE.local.md, in .gitignore | One engineer's notes |
@docs/architecture.md.CLAUDE.md from the repo, then cut it in half. Every time Claude makes the same mistake twice in the pilot, add 1 line.CLAUDE.md tells Claude how your codebase works. Skills tell it how your team does a specific job: release notes, a migration, a PR description, a bug report. Write the job down once, commit it, and every engineer and every Claude session does it the same way.SKILL.md file. Claude reads only the description of each skill at the start of a session, and loads the full instructions when a task matches. Engineers can also run one directly with /skill-name (docs).Loading editor...
--- name: release-notes description: Writes customer-facing release notes from merged pull requests. Use when the user asks for release notes, a changelog or what shipped this week. allowed-tools: Read Grep --- 1. List the pull requests merged since the last release tag. 2. Group them into New, Improved and Fixed. Skip refactors and chores. 3. Write 1 line per change for customers, not engineers: what they can do now. 4. Add the notes to CHANGELOG.md under today's date and show the diff.
description does the work. Write it the way an engineer would ask: "Use when the user asks for..." A vague description means Claude never picks the skill up on its own.| Scope | Path | Who gets it |
|---|---|---|
| Project | .claude/skills/, committed | Everyone who works in the repo |
| Personal | ~/.claude/skills/ | 1 engineer, in every project |
| Plugin | | Every repo where the plugin is installed |
| Organization | .claude/skills/ inside the managed settings folder | Every laptop with your managed settings |
| Skill | Triggered by | Saves |
|---|---|---|
release-notes | "What shipped this week?" | The Friday hour someone spends reading merged PRs |
pr-description | "Open a PR for this" | Reviewers asking "what does this change?" |
add-endpoint | "Add an API endpoint for..." | New endpoints that ignore your validation, auth and test conventions |
db-migration | "Add a column to..." | Migrations without a rollback or a backfill |
bug-triage | "Look into this Sentry error" | Fixes before anyone has reproduced the bug |
Loading editor...
{
"extraKnownMarketplaces": {
"acme-tools": {
"source": { "source": "github", "repo": "acme/claude-tools" },
"autoUpdate": true
}
},
"enabledPlugins": {
"team-skills@acme-tools": true
}
}/install-github-app in Claude Code. It installs the GitHub App, adds the secret and opens a workflow PR.@claude in a PR or issue comment to get a review, a fix or an answer.--max-turns and a workflow timeout.| Option | Runs on | Start it with | Use it for |
|---|---|---|---|
/loop | Your laptop, while the session is open | /loop 5m check if the deployment finished | Watching a deploy or a long job. Expires after 7 days |
| Cloud scheduled tasks | Anthropic's cloud, laptop closed | /schedule daily PR review at 9am | Personal routines: a morning PR review, a weekly backlog cleanup |
| GitHub Actions on a schedule | GitHub runners | A workflow with on: schedule | Team jobs that belong to the repo, not to a person |
| Headless claude -p | Any server or CI with cron | claude -p "..." --permission-prompts none | Scripts in your own infrastructure |
/schedule or at claude.ai/code/routines, and pick a trigger: a schedule, an API call or a GitHub event. Two catches from the docs: they're in research preview, and they belong to 1 person's account. They aren't shared with teammates, and their runs count against that person's usage.Loading editor...
name: Morning PR digest
on:
schedule:
- cron: "0 7 * * 1-5" # 7:00 UTC, weekdays
jobs:
digest:
runs-on: ubuntu-latest
steps:
- uses: actions/checkout@v4
- uses: anthropics/claude-code-action@v1
with:
anthropic_api_key: ${{ secrets.ANTHROPIC_API_KEY }}
prompt: "List open pull requests waiting more than 24 hours for review, with author and size, and summarize what merged yesterday."
claude_args: "--max-turns 10"| Automation | Schedule | What it does |
|---|---|---|
| PR digest | Weekdays, 9am | Lists PRs stuck in review and what merged yesterday |
| Alert triage | Every morning | Reads new production errors, groups them and drafts a fix for the top one |
| Dependency updates | Mondays | Bumps minor versions, runs the tests, opens 1 PR |
| Docs drift | Nightly | Flags docs and CLAUDE.md lines that no longer match the code |
| Release notes | Fridays | Runs your release-notes skill and opens a PR with the changelog |
--max-turns, a workflow timeout, and output as a pull request or a comment, never a direct push to main.env block, so engineers can't switch it off per repo (docs):Loading editor...
{
"env": {
"CLAUDE_CODE_ENABLE_TELEMETRY": "1",
"OTEL_METRICS_EXPORTER": "otlp"
}
}| Track | Tells you | Watch for |
|---|---|---|
| Usage: sessions, cost per user | Who adopted it | Engineers who stopped after week 1: ask why |
| Delivery: time to merge, review wait, PRs sent back, releases | Whether the team ships faster | More PRs but longer review waits |
| If | Then |
|---|---|
| Time to merge is down and PRs sent back are flat or down | Roll out to everyone, with the pilot engineers as champions |
| More PRs, but review waits went up | Fix review first: smaller PRs, Claude review on every PR, then expand |
| Pilot engineers stopped using it | Ask them why before buying 50 more seats |
| Option | Best for |
|---|---|
| Do it yourself with this plan and Anthropic's docs | A platform or DevEx lead with time to own it |
| An Anthropic services partner such as Accenture, Deloitte, PwC or KPMG (Claude Partner Network) | Large enterprises with procurement, compliance and thousands of seats |
| Paralect, a specialist team | Teams of 3-25 developers who want the setup done in weeks, not quarters |
| Step | What you get | Price | Time |
|---|---|---|---|
| 1. Report and AI review | Time to merge, review wait and rework from GitHub, AI review on up to 3 repos, agent instructions for your main repo | $490 | 3 days |
| 2. Team rollout | Skills for your 5 most repeated tasks, agent-written tests on every PR, 2 pairing sessions per developer, for up to 10 developers | $2,500 | 2 weeks |
| 3. Whole process | Agents across planning, review, testing, releases and error fixes, for up to 25 developers | $4,900 | 4 weeks |
Read(./.env) and your secrets folders in managed settings, and keep production credentials off developer laptops.@claude mentions on github.com repositories./loop for short checks while a session is open, /schedule for cloud tasks that run with your laptop closed, or a GitHub Actions workflow with a cron schedule for jobs the whole team relies on..claude/skills/ in each repo. For many repos, package them as a plugin in a private marketplace and enable it for everyone from managed settings.Focus on customers and sales while we handle product delivery. Hire a dedicated AI maker or a whole product team.